Current Sensor Fault Detection and Identification for PMSM Drives Using Multichannel Global Maximum Pooling CNN

Abstract

The permanent magnet synchronous motors (PMSM) have been widely used in motor-drive applications. In the closed-loop control system, the current sensors play an essential role, which are highly prone to failures due to high electron-thermal pressure and abnormal vibration. In this article, an efficient current sensor fault detection and identification (FDI) method is proposed for the PMSM drives based on a modified one-dimensional convolutional neural network. In this method, the multichannel input and global maximum pooling layer are dedicatedly designed to achieve a higher diagnostic accuracy. Besides, the number of network parameters is significantly decreased, thereby reducing the training time. Moreover, the data normalization is adopted to enhance the training performance and generalization ability of the proposed network. The effectiveness and generalization of the proposed FDI method are validated via the experimental data obtained on a three-level active neutral-point-clamped-inverter-fed PMSM-drive platform.